Battery cover seal assembly detection method, device, equipment and storage medium

Abstract

The present application relates to the field of image processing technology, and discloses a battery cover seal assembly detection method, device, equipment and storage medium. The method includes: obtaining an initial battery cover image and performing denoising and edge enhancement processing to obtain an edge-enhanced cover image; performing adaptive gamma correction and dynamic range stretching to obtain a brightness-enhanced cover image; performing color judgment to obtain a color judgment result and performing color compensation to obtain a white-balanced cover image; calculating significance weight data and saturation weight data and performing linear weighted fusion to obtain a weighted fusion cover image; obtaining battery cover parameter data and calculating the maximum information coefficient and Pearson correlation coefficient to obtain correlation seal parameter data; performing residual calculation and abnormality judgment through a convolutional neural network model to obtain a seal abnormality judgment result. The present application improves the accuracy of battery cover seal assembly detection.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and in particular to a battery cover sealing assembly detection method, device, equipment and storage medium. Background Art

[0002] As an important energy storage device, batteries are widely used in various electronic devices and new energy vehicles. With the rapid development of the battery industry, the stability and consistency of battery production quality are crucial to ensure battery performance and safety. Among them, the sealing assembly of the battery cover is a key link in the battery production process, which directly affects the sealing performance and service life of the battery.

[0003] At present, the battery cover seal assembly inspection mainly relies on manual visual inspection, which has problems such as low efficiency, strong subjectivity, and easy omissions. In addition, due to the complex surface features of the battery cover, traditional image processing methods are difficult to effectively extract key features, resulting in low detection accuracy. At the same time, the battery cover seal assembly involves multiple process parameters, and the correlation between the parameters has a significant impact on the sealing quality, but the existing methods lack a comprehensive analysis of these parameters, so the accuracy of the existing technology is low. Summary of the invention

[0004] The present application provides a battery cover sealing assembly detection method, device, equipment and storage medium, which are used to improve the accuracy of battery cover sealing assembly detection.

[0005] In a first aspect, the present application provides a battery cover sealing assembly detection method, the battery cover sealing assembly detection method comprising:

[0006] Acquire an initial battery cover image and perform denoising and edge enhancement processing to obtain an edge-enhanced cover image;

[0007] Performing adaptive gamma correction and dynamic range stretching on the edge-enhanced cover image to obtain a brightness-enhanced cover image;

[0008] Performing color judgment on the brightness enhanced cover plate image to obtain a color judgment result and performing color compensation to obtain a white balanced cover plate image;

[0009] Calculating the saliency weight data of the brightness enhanced cover image and the saturation weight data of the white balanced cover image and performing linear weighted fusion to obtain a weighted fused cover image;

[0010] Obtain the battery cover parameter data and calculate the maximum information coefficient and Pearson correlation coefficient. At the same time, perform correlation analysis to obtain the correlation sealing parameter data.

[0011] The weighted fused cover plate image and the correlation sealing parameter data are input into a preset convolutional neural network model for residual calculation and abnormality determination to obtain a sealing abnormality determination result.

[0012] In a second aspect, the present application provides a battery cover sealing assembly detection device, the battery cover sealing assembly detection device comprising:

[0013] An acquisition module is used to acquire an initial battery cover image and perform denoising and edge enhancement processing to obtain an edge-enhanced cover image;

[0014] A correction module, used for performing adaptive gamma correction and dynamic range stretching on the edge-enhanced cover image to obtain a brightness-enhanced cover image;

[0015] A compensation module, used to perform color judgment on the brightness enhanced cover image, obtain a color judgment result and perform color compensation to obtain a white balanced cover image;

[0016] A fusion module, used for calculating the saliency weight data of the brightness enhancement cover image and the saturation weight data of the white balance cover image and performing linear weighted fusion to obtain a weighted fused cover image;

[0017] A calculation module is used to obtain the battery cover parameter data and calculate the maximum information coefficient and Pearson correlation coefficient, and at the same time, perform correlation analysis to obtain correlation sealing parameter data;

[0018] The determination module is used to input the weighted fusion cover plate image and the correlation sealing parameter data into a preset convolutional neural network model to perform residual calculation and abnormality determination to obtain a sealing abnormality determination result.

[0019] The third aspect of the present application provides a battery cover sealing assembly detection device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the battery cover sealing assembly detection device performs the above-mentioned battery cover sealing assembly detection method.

[0020] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned battery cover sealing assembly detection method.

[0021] In the technical solution provided by the present application, the present application adopts a variety of image enhancement technologies, such as edge enhancement, adaptive gamma correction, dynamic range stretching and color compensation, which can effectively improve the contrast, brightness and color reproduction of the image, and improve the quality and accuracy of image processing. Not only the significant features of the image are considered, but also the saturation features are integrated, so that the feature expression is more comprehensive and the key information of the battery cover image can be better captured. By calculating the maximum information coefficient and the Pearson correlation coefficient and performing principal component analysis, the present application can effectively mine the intrinsic correlation between multiple sealing parameters and provide an important basis for abnormal determination. The convolutional neural network model is adopted to improve the intelligent level of abnormal detection. The present application inputs the weighted fused image features and the correlation sealing parameter data into the preset convolutional neural network model, and uses the powerful feature learning ability of the model to more accurately calculate the residual and determine the abnormal area. Automated and intelligent detection is realized. Compared with traditional manual visual inspection, the present application realizes the automation and intelligence of battery cover sealing assembly detection through the organic combination of image processing, feature fusion, correlation analysis and convolutional neural network model, and improves the detection efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0023] Figure 1 A schematic diagram of an embodiment of a battery cover sealing assembly detection method in an embodiment of the present application;

[0024] Figure 2 Schematic diagram of an embodiment of a battery cover sealing assembly detection device in an embodiment of the present application. DETAILED DESCRIPTION

[0025] Embodiments of the present application provide a battery cover seal assembly detection method, device, equipment and storage medium. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiment of the present application, an embodiment of the battery cover sealing assembly detection method includes:

[0027] Step S101, obtaining an initial battery cover image and performing denoising and edge enhancement processing to obtain an edge enhanced cover image;

[0028] It is understandable that the execution subject of the present application may be a battery cover sealing assembly detection device, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0029] Specifically, an initial battery cover image is obtained, and pixel analysis is performed on the image to obtain pixel noise data. Based on the noise data, the initial battery cover image is subjected to median filtering to remove noise in the image and obtain a filtered cover image. The filtered cover image is subjected to gradient calculation to generate a gradient image. The gradient image is subjected to non-maximum suppression to obtain a non-maximum suppression image, and the edge features of the image are enhanced to make the edge information clearer. The non-maximum suppression image is subjected to double threshold processing to obtain a double threshold processing image. Through double threshold processing, significant edge information in the image is extracted. The double threshold processing image is subjected to edge connection to obtain an edge connection image. The continuity of the edge is ensured, so that the edge features in the image are more complete and obvious. The edge connection image is used to extract the edge of the initial battery cover image to obtain an edge extraction image. The edge extraction image is subjected to histogram equalization processing to obtain an equalized image. Histogram equalization can adjust the grayscale distribution of the image and enhance the overall contrast of the image. On this basis, the Laplacian operator of the equalized image is calculated to enhance the detail features of the image. The image is Gaussian smoothed using the Laplace operator to obtain a smoothed image. Gaussian smoothing can effectively reduce the noise in the image while maintaining the edge information of the image. The smoothed image is nonlinearly expanded to obtain an expanded image. Nonlinear expansion can enhance the local features of the image and make the important information in the image more prominent. The expanded image is weightedly fused with the initial battery cover image at the pixel level to generate an edge-enhanced cover image.

[0030] Step S102, performing adaptive gamma correction and dynamic range stretching on the edge-enhanced cover image to obtain a brightness-enhanced cover image;

[0031] Specifically, pixel statistics are performed on the edge-enhanced cover image to generate a pixel statistics histogram. By analyzing the histogram, the pixel distribution of the image is obtained. Based on the pixel statistics histogram, the cumulative distribution function is calculated to further process the brightness information of the image. Gaussian fitting is performed through the cumulative distribution function to obtain Gaussian fitting parameters. Using these parameters, an adaptive gamma correction coefficient is calculated, and the coefficient is used to adjust the gamma value of the image, thereby realizing the brightness correction of the image. Gamma transformation is performed on the edge-enhanced cover image to generate a gamma transformation image. Gamma transformation can effectively adjust the brightness distribution of the image, so that the contrast of the image is enhanced. Histogram normalization is performed on the gamma transformation image to obtain a normalized image. Histogram normalization can make the grayscale distribution of the image more uniform, thereby improving the visual effect of the image. Based on the normalized image, its logarithmic mean is calculated to obtain logarithmic mean data. Through these data, a dynamic range stretching coefficient is calculated, and the coefficient is used to adjust the dynamic range of the image, so that the brightness level of the image is richer. The dynamic range stretching coefficient is used to stretch the gamma transformation image to obtain a stretched image. Dynamic range stretching can expand the grayscale range of an image, making the details in the image clearer. The stretched image is weighted fused with the gamma transform image at the pixel level to generate a brightness enhanced cover image. Combining the advantages of the stretched image and the gamma transform image, the final brightness enhanced cover image has significant improvements in brightness and contrast.

[0032] Step S103, performing color judgment on the brightness enhanced cover image, obtaining a color judgment result and performing color compensation to obtain a white balanced cover image;

[0033] Specifically, pixel segmentation is performed on the brightness enhanced cover plate image to obtain a segmented image. Through pixel segmentation, the image is divided into several regions, and the color characteristics of each region are analyzed more accurately. The RGB component histogram is calculated based on the segmented image to obtain RGB histogram data. The RGB histogram data reflects the distribution of the three color components of red, green and blue in the image. The RGB histogram data is fitted with a Gaussian mixture model to obtain Gaussian mixture model parameters. The Gaussian mixture model can effectively describe the characteristics of color distribution, and the main statistical characteristics of the RGB histogram are obtained through the fitting process. According to the Gaussian mixture model parameters, the RGB histogram data is color classified to obtain a color classification result. The color classification result is subjected to main color extraction to obtain main color data. The main color features of the image are identified from the color classification result, and these main color data reflect the overall color trend of the image. Based on the main color data, color detection is performed on the brightness enhanced cover plate image to obtain a color detection result. Threshold judgment is performed on the color detection result to obtain a threshold judgment result. The color detection result is compared with a preset color standard to determine whether the image meets a specific color requirement. According to the threshold judgment result, the corresponding color judgment result is determined. The color judgment result reflects whether the color of the image needs to be adjusted. The color gain coefficient is calculated based on the color judgment result. The color gain coefficient is used to adjust the color components of the image to achieve color compensation. The edge enhancement cover image is color-gained by the color gain coefficient to obtain a white balance cover image. The color gain process refers to adjusting the gain of the red, green, and blue color components in the image according to the calculated color gain coefficient, so that the color of the image is more balanced and natural.

[0034] Step S104, calculating the saliency weight data of the brightness enhanced cover image and the saturation weight data of the white balanced cover image and performing linear weighted fusion to obtain a weighted fused cover image;

[0035] Specifically, the brightness enhanced cover image is decomposed by Gaussian pyramid to obtain a Gaussian pyramid image set. Gaussian pyramid decomposition is a multi-scale image processing technology that extracts multi-scale features of the image by performing multi-level smoothing and downsampling on the image. The central annular contrast is calculated based on the Gaussian pyramid image set to obtain a contrast image. The central annular contrast is an indicator for measuring the local contrast of an image, which can highlight the salient area in the image. The contrast image is distance transformed to obtain a distance transformed image. Distance transformation is an image processing technology used to calculate the distance from each pixel in the image to the nearest salient area to generate a distance transformed image. Based on the distance transformed image, a saliency map is calculated to obtain a saliency image. The saliency map is an image representing the salient area in the image. By calculating the saliency value of each pixel, the important area in the image is highlighted. The initial saliency image is normalized to obtain saliency weight data. Normalization can adjust the saliency value to a standard range for subsequent processing. At the same time, the white balance cover image is converted to HSV color space to obtain an HSV image. The HSV color space is a color representation method that is more in line with human visual perception, where H represents hue, S represents saturation, and V represents brightness. According to the HSV image, the saturation component is extracted to obtain a saturation image. The saturation component reflects the purity of the color in the image and is an important indicator for color analysis. The saturation image is normalized to obtain saturation weight data. According to the saliency weight data and the saturation weight data, the brightness enhancement cover image and the white balance cover image are linearly weighted fused to obtain a weighted fused cover image. By weighted averaging different images according to certain weights, the advantages of different images are combined to generate a fused image with higher quality.

[0036] Step S105, obtaining battery cover parameter data and calculating the maximum information coefficient and Pearson correlation coefficient, and at the same time, performing correlation analysis to obtain correlation sealing parameter data;

[0037] Specifically, the battery cover parameter data is obtained, and the battery cover parameter data is standardized to obtain standardized parameter data. Standardization can eliminate the dimensional differences between different parameters, so that the parameters can be compared on the same scale. The covariance matrix is ​​calculated according to the standardized parameter data to obtain the covariance matrix data. The covariance matrix is ​​used to describe the linear relationship between the parameters. The covariance matrix data is subjected to eigenvalue decomposition to obtain eigenvalues ​​and eigenvectors. The eigenvalues ​​and eigenvectors reflect the variance and weight of the data in each principal component direction. The cumulative contribution rate is calculated according to the eigenvalues ​​and eigenvectors to obtain the cumulative contribution rate data. The cumulative contribution rate indicates the contribution of each principal component to the overall variance of the data. The cumulative contribution rate data is threshold screened to determine the number of characteristic roots to be retained to obtain the number of screened characteristic roots. By screening the number of characteristic roots, the principal component that best represents the data characteristics is selected. According to the number of screened characteristic roots, the principal component is extracted from the standardized parameter data to obtain the principal component data. The principal component is a linear combination of the original parameters, which can retain the variance information of the data to the greatest extent. The maximum information coefficient is calculated for the principal component data to obtain the maximum information coefficient. The maximum information coefficient is an indicator to measure the nonlinear relationship between variables, which can reveal the complex relationship between variables. At the same time, the Pearson correlation coefficient is calculated for the principal component data to obtain the Pearson correlation coefficient. The Pearson correlation coefficient is used to measure the linear relationship between two variables. The value range is between -1 and 1, reflecting the degree of positive or negative correlation between the variables. According to the maximum information coefficient and the Pearson correlation coefficient, a correlation scoring matrix is ​​constructed. The correlation scoring matrix combines the advantages of the maximum information coefficient and the Pearson correlation coefficient, and can comprehensively describe the correlation between variables. The correlation scoring matrix is ​​sorted to obtain the correlation sealing parameter data. The sorting process can identify the parameters that have the greatest impact on the sealing performance of the battery cover, thereby providing an important basis for sealing assembly detection.

[0038] Step S106: input the weighted fused cover plate image and the correlation sealing parameter data into a preset convolutional neural network model for residual calculation and abnormality determination to obtain a sealing abnormality determination result.

[0039] Specifically, the weighted fusion cover image is scaled to obtain a scaled image. Scale transformation can adjust the size of the image to adapt it to the input requirements of the convolutional neural network model. The scaled image is input into the encoder in the preset convolutional neural network model for feature encoding processing to obtain a coded feature map. The encoder extracts high-dimensional features of the image through multi-layer convolution operations, and the generated coded feature map contains rich image information. The coded feature map is input into the decoder in the convolutional neural network model for feature decoding processing to obtain a decoded feature map. The decoder restores the coded feature map to a feature map of the same size as the input image through layer-by-layer deconvolution operations. The decoded feature map is upsampled to obtain an upsampled feature map. The upsampling process improves the resolution of the feature map through interpolation or deconvolution operations to make it consistent with the size of the scaled image. The upsampled feature map is subtracted element by element from the scaled image to obtain a residual image. The residual image reflects the part that cannot be accurately restored in the image reconstruction process. By analyzing the residual image, the abnormal area in the image is identified. The residual image is pixel clustered to obtain a clustered image. Pixels with similar features are grouped together to highlight abnormal areas in the image. The abnormal area is extracted based on the clustered image to obtain an abnormal area image. The abnormal area image identifies areas where there may be sealing problems. At the same time, the correlation sealing parameter data is input into the classifier in the convolutional neural network model for parameter classification and analysis to obtain classification results. The classifier analyzes the correlation sealing parameter data to identify parameter features that may cause sealing abnormalities. Anomaly judgment is performed based on the abnormal area image and the classification results to obtain a sealing abnormality judgment result. The abnormality judgment process combines the analysis results of the image and parameters to accurately identify abnormal conditions in the sealing assembly.

[0040] In the embodiment of the present application, the present application adopts a variety of image enhancement technologies, such as edge enhancement, adaptive gamma correction, dynamic range stretching and color compensation, which can effectively improve the contrast, brightness and color reproduction of the image, and improve the quality and accuracy of image processing. Not only the significant features of the image are considered, but also the saturation features are integrated, so that the feature expression is more comprehensive and the key information of the battery cover image can be better captured. By calculating the maximum information coefficient and the Pearson correlation coefficient and performing principal component analysis, the present application can effectively mine the intrinsic correlation between multiple sealing parameters and provide an important basis for abnormal determination. The convolutional neural network model is adopted to improve the intelligent level of abnormal detection. The present application inputs the weighted fused image features and the correlation sealing parameter data into the preset convolutional neural network model, and uses the powerful feature learning ability of the model to more accurately calculate the residual and determine the abnormal area. Automated and intelligent detection is realized. Compared with traditional manual visual inspection, the present application realizes the automation and intelligence of battery cover sealing assembly detection through the organic combination of image processing, feature fusion, correlation analysis and convolutional neural network model, and improves the detection efficiency and accuracy.

[0041] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0042] (1) obtaining an initial battery cover image, and performing pixel analysis on the initial battery cover image to obtain pixel noise data;

[0043] (2) performing median filtering on the initial battery cover image according to the pixel noise data to obtain a filtered cover image;

[0044] (3) calculating a gradient image based on the filtered cover image, and performing non-maximum suppression on the gradient image to obtain a non-maximum suppressed image;

[0045] (4) performing double threshold processing on the non-maximum suppression image to obtain a double threshold processed image, and performing edge connection on the double threshold processed image to obtain an edge connected image;

[0046] (5) performing edge extraction on the initial battery cover plate image according to the edge connection image to obtain an edge extraction image;

[0047] (6) performing histogram equalization on the edge extraction image to obtain an equalized image, and calculating the Laplace operator based on the equalized image;

[0048] (7) performing Gaussian smoothing using a Laplace operator to obtain a smoothed image, and performing nonlinear expansion on the smoothed image to obtain an expanded image;

[0049] (8) Perform pixel-level weighted fusion on the expanded image and the initial battery cover image to obtain an edge-enhanced cover image.

[0050] Specifically, an initial battery cover image is obtained, and pixel analysis is performed on it to obtain pixel noise data. The purpose of pixel analysis is to identify and quantify the noise level in the image so that effective denoising can be performed in subsequent steps. Assume that the initial image is I(x, y), where x and y are the coordinates of the image, and the noise data can be expressed as N(x, y). Median filtering is performed on the initial battery cover image according to the pixel noise data to obtain a filtered cover image. Median filtering is a commonly used denoising technique that can effectively remove salt and pepper noise while retaining the edge information of the image. The mathematical representation of median filtering is:

[0051] I med (x,y)=median{I(i,j)|(i,j)∈W(x,y)}

[0052] Among them, I med (x, y) is the filtered image, W(x, y) is the window centered at (x, y), and median means taking the median of all pixel values ​​in the window. The gradient image is calculated based on the filtered cover image, and the gradient image is non-maximum suppressed to obtain a non-maximum suppressed image. The gradient image can be calculated using the Sobel operator to calculate the horizontal and vertical gradients of the image.

[0053] G x =I med *S x , G y =I med *S y ;

[0054] Among them, G x and G y Respectively represent the gradient of the image in the horizontal and vertical directions, * represents the convolution operation, S x and S y are the horizontal and vertical kernels of the Sobel operator respectively. The gradient magnitude can be expressed as:

[0055]

[0056] Non-maximum suppression suppresses non-maximum pixels in the gradient amplitude image and retains pixels that may belong to the edge to obtain a non-maximum suppressed image. Double threshold processing is performed on the non-maximum suppressed image to obtain a double threshold processed image. Double threshold processing uses two thresholds to divide the gradient amplitude image into three categories: strong edge, weak edge and non-edge. Through edge connection, the weak edge is connected to the strong edge to obtain a complete edge connection image. Based on the edge connection image, the initial battery cover image is edge extracted to obtain an edge extracted image. The edge features are identified and separated from the original image. The edge extracted image is subjected to histogram equalization to obtain an equalized image. Histogram equalization improves the contrast of the image by stretching the grayscale distribution of the image, making the details more obvious. Based on the equalized image, the Laplace operator is calculated. The Laplace operator is the second-order derivative of the image and is used to detect the edges in the image. The calculation formula of the Laplace operator is:

[0057]

[0058] Where ΔI represents the Laplace operator, and are the second-order derivatives of the image in the x and y directions respectively. The image is Gaussian smoothed by the Laplace operator to obtain a smoothed image. The purpose of Gaussian smoothing is to reduce the noise in the image while retaining the edge information. The mathematical representation of Gaussian smoothing is:

[0059] I smooth =I*G σ

[0060] Among them, G σ is a Gaussian kernel with standard deviation σ. The smoothed image is subjected to nonlinear dilation to obtain a dilated image. Nonlinear dilation is a morphological operation that enhances local features in an image and makes edges more prominent. The dilation operation can be expressed as:

[0061]

[0062] Among them, I dil (x, y) is the expanded image, and S is the structural element. The expanded image is weighted fused with the initial battery cover image at the pixel level to obtain an edge-enhanced cover image. Combining the edge information of the expanded image with the original details of the initial image, an image with high contrast and clear edges is generated. The formula for weighted fusion is:

[0063] I enh (x,y)=α·I dil (x,y)+(1-α)·I(x,y);

[0064] Among them, I enh(x, y) is the edge-enhanced cover image, and α is the fusion coefficient, which controls the weighted ratio of the two images.

[0065] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0066] (1) Performing pixel statistics on the edge enhanced cover image to obtain a pixel statistical histogram, and calculating a cumulative distribution function based on the pixel statistical histogram;

[0067] (2) performing Gaussian fitting through the cumulative distribution function to obtain Gaussian fitting parameters, and calculating the adaptive gamma correction coefficient according to the Gaussian fitting parameters;

[0068] (3) performing gamma transformation on the edge enhanced cover plate image to obtain a gamma transformation image, and performing histogram normalization on the gamma transformation image to obtain a normalized image;

[0069] (4) calculating the logarithmic mean according to the normalized image to obtain logarithmic mean data, and calculating the dynamic range stretching coefficient according to the logarithmic mean data;

[0070] (5) performing dynamic range stretching on the gamma transformed image using a dynamic range stretching coefficient to obtain a stretched image;

[0071] (6) Perform pixel-level weighted fusion of the stretched image and the gamma transformed image to obtain a brightness enhanced cover image.

[0072] Specifically, pixel statistics are performed on the edge enhanced cover image to obtain a pixel statistical histogram. The pixel statistical histogram shows the number of pixels at each gray level in the image, reflecting the overall brightness distribution of the image. The cumulative distribution function F(x) is calculated based on the pixel statistical histogram, which represents the proportion of pixels less than or equal to a certain gray level x, and can be expressed by the formula:

[0073]

[0074] Where h(i) is the number of pixels at gray level i in the histogram, and N is the total number of pixels in the image. Gaussian fitting is performed through the cumulative distribution function to obtain Gaussian fitting parameters. Gaussian fitting is used to approximate the cumulative distribution function to a Gaussian distribution, thereby simplifying the brightness correction process. The mathematical expression of Gaussian distribution is:

[0075]

[0076] Where μ is the mean, σ is the standard deviation, and G(x) is the Gaussian distribution function. Through the fitting process, the two Gaussian fitting parameters μ and σ are obtained. Based on the Gaussian fitting parameters, the adaptive gamma correction coefficient is calculated. Gamma correction is used to adjust the brightness of the image to make the grayscale of the image more uniform. The calculation formula of the gamma correction coefficient γ is

[0077]

[0078] Among them, F mean is the mean of the cumulative distribution function, G mean is the mean of the Gaussian distribution. Perform gamma transformation on the edge-enhanced cover image to obtain a gamma transformed image. The mathematical expression of gamma transformation is:

[0079] I γ (x,y)=I(x,y) γ ;

[0080] Among them, I(x,y) is the pixel value of the original image, I γ (x, y) is the pixel value of the image after gamma transformation, and γ is the gamma correction coefficient. The gamma transformed image is histogram normalized to obtain a normalized image. Histogram normalization is used to adjust the grayscale distribution of the image to make it more uniform and standardized. Through normalization, the overly bright or dark areas of the image are eliminated, thereby improving the overall brightness and contrast of the image. Based on the normalized image, the logarithmic mean is calculated to obtain the logarithmic mean data. The logarithmic mean is used to further adjust the brightness distribution of the image, and its calculation formula is:

[0081]

[0082] Among them, μ log is the logarithmic mean, N is the total number of pixels in the image, and I γ (x i ,y i ) is the pixel value in the gamma transformed image. Based on the logarithmic mean data, the dynamic range stretching coefficient is calculated. Dynamic range stretching is used to expand the grayscale range of the image, making the brightness level of the image richer. The calculation formula of the dynamic range stretching coefficient is:

[0083]

[0084] Among them, α stretch is the dynamic range stretch factor, max(I γ ) and min(I γ ) are the maximum and minimum pixel values ​​of the gamma transform image, respectively. The dynamic range stretching coefficient is used to stretch the gamma transform image to obtain a stretched image. The mathematical expression for dynamic range stretching is:

[0085] I stretch (x,y)=α stretch ·log(I γ (x,y));

[0086] Among them, I stretch (x, y) is the pixel value of the stretched image. The stretched image is weightedly fused with the gamma transformed image at the pixel level to obtain a brightness enhanced cover image. Combining the brightness information of the stretched image with the original details of the gamma transformed image, an image with high contrast and clear details is generated. The mathematical expression of weighted fusion is:

[0087] I enh (x,y)=β·I stretch (x,y)+(1-β)·I γ (x,y);

[0088] Among them, I enh (x, y) is the pixel value of the brightness enhanced cover image, and β is the fusion coefficient, which controls the weighted ratio of the two images.

[0089] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0090] (1) performing pixel segmentation on the brightness enhanced cover plate image to obtain a segmented image, and calculating an RGB component histogram based on the segmented image to obtain RGB histogram data;

[0091] (2) Fitting the Gaussian mixture model to the RGB histogram data to obtain the Gaussian mixture model parameters;

[0092] (3) Color classification is performed on the RGB histogram data according to the Gaussian mixture model parameters to obtain a color classification result;

[0093] (4) extracting the main color tone from the color classification result to obtain main color tone data, and performing color detection on the brightness enhanced cover plate image based on the main color tone data to obtain a color detection result;

[0094] (5) performing a threshold judgment on the color detection result to obtain a threshold judgment result, and determining a corresponding color judgment result according to the threshold judgment result;

[0095] (6) Calculate a color gain coefficient based on the color judgment result, and perform color gain on the edge enhanced cover image using the color gain coefficient to obtain a white balanced cover image.

[0096] Specifically, pixel segmentation is performed on the brightness enhanced cover image to obtain a segmented image. The image is divided into several regions with similar color features. The RGB component histogram is calculated based on the segmented image to obtain RGB histogram data. The RGB histogram data represents the distribution of pixel values ​​of the three color channels of red, green, and blue in the image. A Gaussian mixture model is fitted to the RGB histogram data to obtain Gaussian mixture model parameters. The Gaussian mixture model is a statistical model used to represent the distribution of data. Its mathematical expression is:

[0097]

[0098] Where K is the number of Gaussian distributions, π i is the weight of the ith Gaussian distribution, μ i and are the mean and variance of the i-th Gaussian distribution. Through the fitting process, we get these parameters, namely π i , μ i and The RGB histogram data is color classified according to the Gaussian mixture model parameters to obtain the color classification result. The pixels in the image are divided into different categories according to their color features. Based on the Gaussian mixture model parameters, the color category of each pixel can be determined by maximum likelihood estimation, that is, the pixel is assigned to the Gaussian distribution that maximizes the likelihood function. The color classification result is subjected to main color extraction to obtain main color data. The main color features of the image are identified from the color classification result, and these data reflect the overall color trend of the image. The brightness enhanced cover image is color detected according to the main color data to obtain the color detection result. The color information of the image is compared with the main color data to identify whether there is a color anomaly in the image. The color detection result is threshold judged to obtain the threshold judgment result. The color detection result is compared with the preset color standard to determine whether the color of the image meets the requirements. The threshold judgment can be achieved by setting one or more color thresholds, which are used to determine whether the color is within an acceptable range. According to the threshold judgment result, the corresponding color judgment result is determined. If the color detection result exceeds the preset threshold range, it indicates that the color of the image is abnormal and color compensation is required. According to the color judgment result, the color gain coefficient is calculated. The color gain coefficient is used to adjust the color component of the image to achieve color compensation. The calculation formula of color gain coefficient is:

[0099]

[0100] Among them, G c is the color gain coefficient, M c is the target color mean, A cis the actual color mean. The edge enhanced cover image is color-gained by the color gain coefficient to obtain the white balanced cover image. The mathematical expression of color gain is:

[0101] I wb (x,y,c)=G c I(x,y,c);

[0102] Among them, I wb (x, y, c) is the pixel value of the white balance cover image at position (x, y) and color channel c, G c The color gain coefficient I(x,y,c) is the pixel value of the original image at the same position and color channel. For example, suppose there is a brightness-enhanced cover image with a pixel value range between 0 and 255. Suppose the threshold judgment result shows that a color channel of the image exceeds the preset threshold range. Calculate the color gain coefficient based on the color judgment result. Suppose the target color mean is 128 and the actual color mean is 100, then the color gain coefficient is:

[0103]

[0104] The image is color-gained by the color gain coefficient to obtain the white balance cover image. Assuming that the pixel value of the original image at a certain position and color channel is 100, the pixel value of the white balance cover image at the same position and color channel is:

[0105] I wb (x, y, c) = 1.28 100 = 128;

[0106] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0107] (1) Performing Gaussian pyramid decomposition on the brightness enhanced cover image to obtain a Gaussian pyramid image set;

[0108] (2) calculating the central annular contrast according to the Gaussian pyramid image set to obtain a contrast image, and performing distance transformation on the contrast image to obtain a distance transformation image;

[0109] (3) Calculating the saliency map according to the distance transformation image to obtain a saliency image, and normalizing the initial saliency image to obtain saliency weight data;

[0110] (4) performing HSV color space conversion on the white balance cover image to obtain an HSV image, and extracting a saturation component from the HSV image to obtain a saturation image;

[0111] (5) Normalizing the saturation image to obtain saturation weight data;

[0112] (6) According to the saliency weight data and the saturation weight data, a linear weighted fusion is performed on the brightness enhanced cover image and the white balanced cover image to obtain a weighted fused cover image.

[0113] Specifically, the brightness enhanced cover image is decomposed by Gaussian pyramid to obtain a Gaussian pyramid image set. Gaussian pyramid decomposition is a multi-scale image processing technology that generates multiple image layers with different resolutions by performing step-by-step Gaussian blurring and downsampling on the image. The central annular contrast is calculated based on the Gaussian pyramid image set to obtain a contrast image. The central annular contrast is a method to measure the local contrast of an image, which is calculated by comparing the difference between the central pixel and the surrounding pixels. Specifically, for each layer of the image, the calculation formula is:

[0114]

[0115] Where C(x,y) is the contrast value, Ω is the set of pixels around the central pixel (x,y), and N is the number of surrounding pixels. Perform distance transform on the contrast image to obtain a distance transformed image. Distance transform is an image processing operation used to calculate the distance from each pixel in the image to the nearest salient edge. In this way, a distance transformed image is generated, which represents the distance from each pixel to the nearest edge. Based on the distance transformed image, the saliency map is calculated to obtain a saliency image. The saliency map is used to highlight important areas in the image, and its calculation method is usually based on a comprehensive analysis of distance transform and contrast. The generation of a saliency image can be achieved by performing a nonlinear transformation on the distance transformed image. The calculation formula of the saliency image S(x,y) is:

[0116]

[0117] Where D(x,y) is the distance transformation value, σ is an adjustment parameter, and C(x,y) is the contrast value. The initial saliency image is normalized to obtain saliency weight data. Normalization adjusts the range of saliency values ​​to make it suitable for further processing and analysis. At the same time, the white balance cover image is converted to HSV color space to obtain an HSV image. HSV color space is a color representation method that is more in line with human visual perception, where H represents hue, S represents saturation, and V represents brightness. The color information of the image is extracted through HSV color space conversion. The saturation component is extracted from the HSV image to obtain a saturation image. The saturation image represents the color purity of each pixel in the image and is an important indicator for color analysis. The saturation image is normalized to obtain saturation weight data. According to the saliency weight data and the saturation weight data, the brightness enhancement cover image and the white balance cover image are linearly weighted fused to obtain a weighted fused cover image.

[0118] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0119] (1) Obtaining battery cover plate parameter data, and performing standardization processing on the battery cover plate parameter data to obtain standardized parameter data;

[0120] (2) calculating the covariance matrix according to the standardized parameter data to obtain covariance matrix data, and performing eigenvalue decomposition on the covariance matrix data to obtain eigenvalues ​​and eigenvectors;

[0121] (3) Calculating the cumulative contribution rate according to the eigenvalue and the eigenvector to obtain the cumulative contribution rate data, and performing threshold screening on the cumulative contribution rate data to obtain the number of screening characteristic roots;

[0122] (4) extracting principal components from the standardized parameter data according to the number of screening characteristic roots to obtain principal component data;

[0123] (5) Calculating the maximum information coefficient on the principal component data to obtain the maximum information coefficient, and calculating the Pearson correlation coefficient on the principal component data to obtain the Pearson correlation coefficient;

[0124] (6) Construct a correlation scoring matrix based on the maximum information coefficient and the Pearson correlation coefficient, and sort the correlation scoring matrix to obtain the correlation sealing parameter data.

[0125] Specifically, the battery cover parameter data is obtained and standardized to eliminate the dimensional differences between different parameters so that the parameters can be compared on the same scale. The covariance matrix is ​​calculated based on the standardized parameter data to obtain the covariance matrix data. The covariance matrix is ​​used to describe the linear relationship between the parameters, and its calculation formula is:

[0126]

[0127] Where Σ is the covariance matrix, x i is the i-th standardized data, μ is the mean vector, and n is the number of samples. Perform eigenvalue decomposition on the covariance matrix data to obtain eigenvalues ​​and eigenvectors. The eigenvalues ​​and eigenvectors reflect the variance and weight of the data in each principal component direction. The formula for eigenvalue decomposition is:

[0128] Σv=λv;

[0129] Among them, Σ is the covariance matrix, v is the eigenvector, and λ is the eigenvalue. The cumulative contribution rate is calculated based on the eigenvalue and eigenvector to obtain the cumulative contribution rate data. The cumulative contribution rate indicates the contribution of each principal component to the overall variance of the data. The calculation formula for the cumulative contribution rate is:

[0130]

[0131] Among them, λ i is the i-th eigenvalue, k is the number of selected eigenvalues, and p is the total number of eigenvalues. The cumulative contribution rate data is threshold screened to determine the number of eigenvalues ​​that need to be retained and the number of screened eigenvalues ​​is obtained. By screening the number of eigenvalues, the principal component that best represents the data characteristics is selected. According to the number of screened eigenvalues, the principal components are extracted from the standardized parameter data to obtain the principal component data. The principal component is a linear combination of the original parameters, which can retain the variance information of the data to the greatest extent. The maximum information coefficient is calculated for the principal component data to obtain the maximum information coefficient. The maximum information coefficient is an indicator for measuring the nonlinear relationship between variables, which can reveal the complex relationship between variables. At the same time, the Pearson correlation coefficient is calculated for the principal component data to obtain the Pearson correlation coefficient. The Pearson correlation coefficient is used to measure the linear relationship between two variables, and its calculation formula is:

[0132]

[0133] Where r is the Pearson correlation coefficient, x i and i are the values ​​of the two variables, and are the means of the two variables, respectively, and n is the number of samples. According to the maximum information coefficient and the Pearson correlation coefficient, a correlation scoring matrix is ​​constructed. The correlation scoring matrix combines the advantages of the maximum information coefficient and the Pearson correlation coefficient, and can fully describe the correlation between variables. The correlation scoring matrix is ​​sorted to obtain the correlation sealing parameter data. The sorting process can identify the parameters that have the greatest impact on the sealing performance of the battery cover, providing an important basis for sealing assembly detection. For example, suppose there is a set of battery cover parameter data, including four variables: pressure, temperature, humidity, and sealing time. These data are standardized to eliminate the influence of different dimensions. The covariance matrix is ​​calculated based on the standardized data, and the eigenvalue decomposition is performed on it to obtain the eigenvalues ​​and eigenvectors. Assuming that the cumulative contribution rate of the first two eigenvalues ​​reaches 90%, two eigenvalues ​​are selected. Based on the selected eigenvalues, two principal components are extracted from the standardized data. The maximum information coefficient and the Pearson correlation coefficient are calculated for the two principal component data respectively. Assume that the calculation results of the maximum information coefficient and the Pearson correlation coefficient show that there is a strong nonlinear and linear relationship between pressure and sealing time. Construct a correlation score matrix, consider the maximum information coefficient and Pearson correlation coefficient, and get the correlation score of each parameter on the sealing performance. Sort the correlation scores to identify the parameters that have the greatest impact on the sealing performance. Assume that the sorting results show that pressure and sealing time are the two key parameters that affect the sealing performance.

[0134] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0135] (1) performing a scale transformation on the weighted fusion cover plate image to obtain a scale transformed image;

[0136] (2) inputting the scale-transformed image into an encoder in a preset convolutional neural network model for feature encoding processing to obtain a coding feature map;

[0137] (3) Inputting the encoded feature map into the decoder in the convolutional neural network model for feature decoding processing to obtain a decoded feature map;

[0138] (4) upsampling the decoded feature map to obtain an upsampled feature map, and performing element-by-element subtraction between the upsampled feature map and the scale-transformed image to obtain a residual image;

[0139] (5) clustering pixels of the residual image to obtain a clustered image, and extracting abnormal regions based on the clustered image to obtain an abnormal region image;

[0140] (6) inputting the correlation sealing parameter data into the classifier in the convolutional neural network model to perform parameter classification and analysis to obtain the classification results;

[0141] (7) An abnormality determination is performed based on the abnormal area image and the classification result to obtain a sealing abnormality determination result.

[0142] Specifically, the weighted fused cover image is scaled to obtain a scaled image. Scale transformation is to adjust the size of the image to adapt it to the input requirements of the preset convolutional neural network model. The scaled image is input into the encoder in the preset convolutional neural network model for feature encoding processing to obtain a coded feature map. The encoder extracts high-dimensional features of the image through multi-layer convolution operations, and the generated coded feature map contains rich image information. The mathematical expression of the convolution operation is:

[0143] F i,j =∑ m,n I i+m,j+n ·K m,n ;

[0144] Among them, F i,j is the convolution result, I i,j is the input image, K m,nis the convolution kernel. The encoded feature map is input into the decoder in the convolutional neural network model for feature decoding to obtain a decoded feature map. The decoder restores the encoded feature map to a feature map of the same size as the input image through layer-by-layer deconvolution operations. The decoded feature map retains the main features of the original image. The decoded feature map is upsampled to obtain an upsampled feature map. The upsampling process improves the resolution of the feature map through interpolation or deconvolution operations to make it consistent with the size of the scaled image. The mathematical expression of upsampling is:

[0145] F′ i,j =F i / α,j / α ;

[0146] Among them, F′ i,jis the upsampled feature map, and α is the upsampling factor. The upsampled feature map is subtracted element by element from the scale-transformed image to obtain a residual image. The residual image reflects the part that cannot be accurately restored during the image reconstruction process. By analyzing the residual image, the abnormal area in the image is identified. The residual image is pixel clustered to obtain a clustered image. The pixel clustering process classifies pixels with similar features into one category to highlight the abnormal area in the image. Common clustering methods include K-means clustering and DBSCAN clustering. Through clustering analysis, a cluster image is generated to represent the distribution of pixels in different categories. The abnormal area is extracted according to the cluster image to obtain the abnormal area image. The abnormal area image identifies the area where there may be sealing problems. The correlation sealing parameter data is input into the classifier in the convolutional neural network model for parameter classification and analysis to obtain the classification result. The classifier analyzes the correlation sealing parameter data and identifies the parameter features that may cause sealing abnormality. Anomaly judgment is performed based on the abnormal area image and the classification result to obtain the sealing abnormality judgment result. The abnormality judgment process combines the analysis results of the image and parameters, and can accurately identify abnormal conditions in the sealing assembly. For example, suppose there is a weighted fused cover image with pixel values ​​ranging from 0 to 255. Scale the image to adjust it to a standard size. For example, resize the image from 1024×1024 to 256×256. Input the scaled image into the encoder in the convolutional neural network model, perform feature encoding, and generate a 64×64 encoded feature map. Input the encoded feature map into the decoder, perform feature decoding, and generate a 256×256 decoded feature map. Upsample the decoded feature map to make its resolution consistent with the scaled image. Assuming the upsampling factor is 4, the upsampled feature map is still 256×256. Subtract the upsampled feature map from the scaled image element by element to obtain the residual image. Assuming the pixel value of the scaled image at a certain position is 150 and the pixel value of the upsampled feature map is 140, the pixel value of the residual image at that position is 10. Pixel clustering is performed on the residual image to obtain a clustered image. For example, using the K-means clustering method, the pixels in the residual image are divided into three categories: normal area, slightly abnormal area and seriously abnormal area. The abnormal area is extracted according to the clustered image, and an abnormal area image is generated. The relevant sealing parameter data (such as temperature, pressure, humidity, etc.) are input into the classifier in the convolutional neural network model for classification and analysis to obtain the classification result. The classification result may show that certain parameters are significantly associated with sealing abnormalities. Finally, the abnormality is judged based on the abnormal area image and the classification result. For example, assuming that the abnormal area image shows that there are serious abnormalities in certain areas, and the classification result shows that high temperature and high pressure are the main factors causing the abnormality, the sealing abnormality judgment result is: under high temperature and high pressure conditions, there is a problem with the sealing of the battery cover.

[0147] The above describes the battery cover seal assembly detection method in the embodiment of the present application. The following describes the battery cover seal assembly detection device in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the battery cover sealing assembly detection device includes:

[0148] An acquisition module 201 is used to acquire an initial battery cover plate image and perform denoising and edge enhancement processing to obtain an edge enhanced cover plate image;

[0149] A correction module 202 is used to perform adaptive gamma correction and dynamic range stretching on the edge-enhanced cover image to obtain a brightness-enhanced cover image;

[0150] The compensation module 203 is used to perform color judgment on the brightness enhanced cover image, obtain a color judgment result and perform color compensation to obtain a white balanced cover image;

[0151] A fusion module 204 is used to calculate the saliency weight data of the brightness enhancement cover image and the saturation weight data of the white balance cover image and perform linear weighted fusion to obtain a weighted fused cover image;

[0152] The calculation module 205 is used to obtain the battery cover parameter data and calculate the maximum information coefficient and the Pearson correlation coefficient, and at the same time, perform correlation analysis to obtain the correlation sealing parameter data;

[0153] The determination module 206 is used to input the weighted fusion cover plate image and the correlation sealing parameter data into a preset convolutional neural network model to perform residual calculation and abnormality determination to obtain a sealing abnormality determination result.

[0154] Through the collaborative cooperation of the above components, the present application adopts a variety of image enhancement technologies, such as edge enhancement, adaptive gamma correction, dynamic range stretching and color compensation, which can effectively improve the contrast, brightness and color reproduction of the image, and improve the quality and accuracy of image processing. Not only the significant features of the image are considered, but also the saturation features are integrated, so that the feature expression is more comprehensive and the key information of the battery cover image can be better captured. By calculating the maximum information coefficient and the Pearson correlation coefficient and performing principal component analysis, the present application can effectively mine the intrinsic correlation between multiple sealing parameters and provide an important basis for abnormal determination. The convolutional neural network model is adopted to improve the intelligent level of abnormal detection. The present application inputs the weighted fused image features and the correlation sealing parameter data into the preset convolutional neural network model, and uses the powerful feature learning ability of the model to more accurately calculate the residual and determine the abnormal area. Automated and intelligent detection is realized. Compared with traditional manual visual inspection, the present application realizes the automation and intelligence of battery cover sealing assembly detection through the organic combination of image processing, feature fusion, correlation analysis and convolutional neural network model, and improves the detection efficiency and accuracy.

[0155] The present application also provides a battery cover sealing assembly detection device, which includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the battery cover sealing assembly detection method in the above-mentioned embodiments.

[0156] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein instructions are stored in the computer, and when the instructions are executed on a computer, the computer executes the steps of the battery cover sealing assembly detection method.

[0157] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0158] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0159] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A battery cover sealing assembly detection method, characterized in that: The battery cover sealing assembly detection method comprises: Acquire an initial battery cover image and perform denoising and edge enhancement processing to obtain an edge-enhanced cover image; Performing adaptive gamma correction and dynamic range stretching on the edge-enhanced cover image to obtain a brightness-enhanced cover image; Performing color judgment on the brightness enhanced cover plate image to obtain a color judgment result and performing color compensation to obtain a white balanced cover plate image; Calculating the saliency weight data of the brightness enhancement cover image and the saturation weight data of the white balance cover image and performing linear weighted fusion to obtain a weighted fused cover image; specifically comprising: performing Gaussian pyramid decomposition on the brightness enhancement cover image to obtain a Gaussian pyramid image set; calculating the central annular contrast according to the Gaussian pyramid image set to obtain a contrast image, and performing distance transformation on the contrast image to obtain a distance transformation image; calculating a saliency map according to the distance transformation image to obtain a saliency image, and normalizing the saliency image to obtain saliency weight data; performing HSV color space conversion on the white balance cover image to obtain an HSV image, and extracting a saturation component according to the HSV image to obtain a saturation image; normalizing the saturation image to obtain saturation weight data; performing linear weighted fusion on the brightness enhancement cover image and the white balance cover image according to the saliency weight data and the saturation weight data to obtain a weighted fused cover image; Obtain the battery cover parameter data and calculate the maximum information coefficient and Pearson correlation coefficient. At the same time, perform correlation analysis to obtain the correlation sealing parameter data. The weighted fused cover plate image and the correlation sealing parameter data are input into a preset convolutional neural network model for residual calculation and abnormality determination to obtain a sealing abnormality determination result.

2. The battery cover sealing assembly detection method according to claim 1, characterized in that: The method of obtaining an initial battery cover plate image and performing denoising and edge enhancement processing to obtain an edge enhanced cover plate image includes: Acquire an initial battery cover image, and perform pixel analysis on the initial battery cover image to obtain pixel noise data; Performing median filtering on the initial battery cover plate image according to the pixel noise data to obtain a filtered cover plate image; Calculating a gradient image according to the filtered cover plate image, and performing non-maximum suppression on the gradient image to obtain a non-maximum suppressed image; Performing double threshold processing on the non-maximum suppression image to obtain a double threshold processed image, and performing edge connection on the double threshold processed image to obtain an edge connected image; Perform edge extraction on the initial battery cover plate image according to the edge connection image to obtain an edge extraction image; Performing histogram equalization on the edge extraction image to obtain an equalized image, and calculating a Laplace operator based on the equalized image; Performing Gaussian smoothing by using the Laplace operator to obtain a smoothed image, and performing nonlinear expansion on the smoothed image to obtain an expanded image; The expanded image is weightedly fused with the initial battery cover image at pixel level to obtain an edge-enhanced cover image.

3. The battery cover sealing assembly detection method according to claim 1, characterized in that: The step of performing adaptive gamma correction and dynamic range stretching on the edge-enhanced cover image to obtain a brightness-enhanced cover image includes: Performing pixel statistics on the edge enhanced cover plate image to obtain a pixel statistical histogram, and calculating a cumulative distribution function according to the pixel statistical histogram; Performing Gaussian fitting through the cumulative distribution function to obtain Gaussian fitting parameters, and calculating adaptive gamma correction coefficients according to the Gaussian fitting parameters; Performing gamma transformation on the edge enhanced cover plate image to obtain a gamma transformed image, and performing histogram normalization on the gamma transformed image to obtain a normalized image; Calculating a logarithmic mean according to the normalized image to obtain logarithmic mean data, and calculating a dynamic range stretch coefficient according to the logarithmic mean data; Performing dynamic range stretching on the gamma transformed image using the dynamic range stretching coefficient to obtain a stretched image; The stretched image and the gamma transformed image are subjected to pixel-level weighted fusion to obtain a brightness enhanced cover plate image.

4. The battery cover sealing assembly detection method according to claim 1, characterized in that: The step of performing color judgment on the brightness enhanced cover plate image, obtaining a color judgment result and performing color compensation to obtain a white balance cover plate image includes: Performing pixel segmentation on the brightness enhanced cover plate image to obtain a segmented image, and calculating an RGB component histogram based on the segmented image to obtain RGB histogram data; Performing Gaussian mixture model fitting on the RGB histogram data to obtain Gaussian mixture model parameters; Performing color classification on the RGB histogram data according to the Gaussian mixture model parameters to obtain a color classification result; Performing main color extraction on the color classification result to obtain main color data, and performing color detection on the brightness enhanced cover plate image according to the main color data to obtain a color detection result; Performing a threshold judgment on the color detection result to obtain a threshold judgment result, and determining a corresponding color judgment result according to the threshold judgment result; A color gain coefficient is calculated according to the color judgment result, and color gain is performed on the edge enhanced cover image using the color gain coefficient to obtain a white balanced cover image.

5. The battery cover sealing assembly detection method according to claim 1, characterized in that: The method of obtaining the battery cover parameter data and calculating the maximum information coefficient and the Pearson correlation coefficient, and performing correlation analysis to obtain the correlation sealing parameter data, includes: Acquiring battery cover plate parameter data, and performing standardization processing on the battery cover plate parameter data to obtain standardized parameter data; Calculating a covariance matrix according to the standardized parameter data to obtain covariance matrix data, and performing eigenvalue decomposition on the covariance matrix data to obtain eigenvalues ​​and eigenvectors; Calculating the cumulative contribution rate according to the eigenvalue and the eigenvector to obtain cumulative contribution rate data, and performing threshold screening on the cumulative contribution rate data to obtain the number of screening characteristic roots; Extracting principal components from the standardized parameter data according to the number of screening characteristic roots to obtain principal component data; Calculating the maximum information coefficient on the principal component data to obtain the maximum information coefficient, and calculating the Pearson correlation coefficient on the principal component data to obtain the Pearson correlation coefficient; A correlation scoring matrix is ​​constructed according to the maximum information coefficient and the Pearson correlation coefficient, and the correlation scoring matrix is ​​sorted to obtain correlation sealing parameter data.

6. The battery cover sealing assembly detection method according to claim 1, characterized in that: The step of inputting the weighted fusion cover plate image and the correlation sealing parameter data into a preset convolutional neural network model for residual calculation and abnormality determination to obtain a sealing abnormality determination result includes: Performing a scale transformation on the weighted fused cover plate image to obtain a scale transformed image; Inputting the scale-transformed image into the encoder in the preset convolutional neural network model for feature encoding processing to obtain a coding feature map; Inputting the encoded feature map into a decoder in the convolutional neural network model for feature decoding processing to obtain a decoded feature map; Upsampling the decoded feature map to obtain an upsampled feature map, and performing element-by-element subtraction between the upsampled feature map and the scale-transformed image to obtain a residual image; Performing pixel clustering on the residual image to obtain a clustered image, and extracting abnormal regions according to the clustered image to obtain an abnormal region image; Inputting the correlation sealing parameter data into the classifier in the convolutional neural network model to perform parameter classification and analysis to obtain a classification result; An abnormality determination is performed based on the abnormal area image and the classification result to obtain a sealing abnormality determination result.

7. A battery cover sealing assembly detection device, characterized in that: The battery cover sealing assembly detection device comprises: An acquisition module is used to acquire an initial battery cover image and perform denoising and edge enhancement processing to obtain an edge-enhanced cover image; A correction module, used for performing adaptive gamma correction and dynamic range stretching on the edge-enhanced cover image to obtain a brightness-enhanced cover image; A compensation module, used to perform color judgment on the brightness enhanced cover plate image, obtain a color judgment result and perform color compensation to obtain a white balanced cover plate image; A fusion module is used to calculate the significance weight data of the brightness enhancement cover image and the saturation weight data of the white balance cover image and perform linear weighted fusion to obtain a weighted fused cover image; specifically comprising: performing Gaussian pyramid decomposition on the brightness enhancement cover image to obtain a Gaussian pyramid image set; calculating the central annular contrast according to the Gaussian pyramid image set to obtain a contrast image, and performing distance transformation on the contrast image to obtain a distance transformation image; calculating a significance map according to the distance transformation image to obtain a significance image, and normalizing the significance image to obtain significance weight data; performing HSV color space conversion on the white balance cover image to obtain an HSV image, and extracting a saturation component according to the HSV image to obtain a saturation image; normalizing the saturation image to obtain saturation weight data; performing linear weighted fusion on the brightness enhancement cover image and the white balance cover image according to the significance weight data and the saturation weight data to obtain a weighted fused cover image; A calculation module is used to obtain the battery cover parameter data and calculate the maximum information coefficient and Pearson correlation coefficient, and at the same time, perform correlation analysis to obtain correlation sealing parameter data; The determination module is used to input the weighted fusion cover plate image and the correlation sealing parameter data into a preset convolutional neural network model to perform residual calculation and abnormality determination to obtain a sealing abnormality determination result.

8. A battery cover sealing assembly detection device, characterized in that: The battery cover sealing assembly detection device comprises: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instructions in the memory to enable the battery cover sealing assembly detection device to execute the battery cover sealing assembly detection method according to any one of claims 1 to 6.

9. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the battery cover sealing assembly detection method according to any one of claims 1 to 6 is implemented.

Images